Evaluating the efficacy of computer aided exercise stress ECG reader (CAESER) for automated diagnosis of coronary artery disease

Primary goal of this project is to significantly elevate the positive predictive value (PPV) and sensitivity of exercise
stress electrocardiography (ESE) in assessing obstructive coronary artery disease (CAD), defined as > 50%
stenosis in at least one major coronary vessel identified by invasive coronary angiography (ICA), across pretest
cardiac risk spectrum using an expert-AI collaborative approach. We leverage a large, harmonized ESE dataset
correlated with ICA of a broad range of patients from three major cardiac care sites in USA (via Mayo Clinic
Integrated Stress Center (MISC)) to develop an automated CAD likelihood determination technique from ESE
with uniform performance across age, sex, race, socio-economic status (SES) and CAD co-morbidities. This will
benefit a key synergistic ongoing effort in wearable electrocardiography (ECG) CAD risk monitoring by this team
via Arizona New Economy Initiative’s (NEI) for a healthy US workforce. Despite lower prevalence, women are
marginally less likely to die from cardiovascular diseases than men and are likely to have poorer prognosis. ICA,
after initial coronary computed tomography angiography (CCTA) triage, is the gold standard for CAD diagnosis,
however, it is expensive and cannot be used frequently on individuals with high SES deprivation index. On the
other hand, CAD risk estimation with clinician review of low-cost non-invasive ESE show poor PPV in men
(77%) and even poorer in women (47%) in population with nearly similar prevalence across sex (0.36 in men
vs. 0.33 in women). An accurate unbiased automated CAD risk determination method from ESE can potentially
pave the way for continuous CAD risk assessment through mobile monitoring using wearable ECG for a broad
range of clinical and general population. The ASU-Mayo research team have developed CAESER (Computer
Aided Exercise Stress ECG Reader), that can automatically analyze ESE and provide likelihood of CAD.
CAESER adopts precision cardiology approach, where baseline ST depression is captured by a personalized
digital twin, which is integrated to continuous CAD risk assessment through a transformer based deep
learning architecture. The first aim is to evaluate the performance of CAESER as compared to ICA
evidence of CAD across large scale symptomatic patient population. CAESER will be tested on 40,000 ICA
correlated ESE cases from the MISC repository as well as independent secondary dataset from Physionet. The
second aim is to evaluate the variance in performance of CAESER across gender, race and SES. The
performance of CAESER developed in AIM 1 will be evaluated for variance across sex, race and area
deprivation index (ADI) using metrics such as precision, sensitivity, specificity, area under the curve of ROC.
Using a multi-variate logistic regression model, the variance of CAESER performance metrics with respect to
the ADI percentile of patients will be evaluated for statistically significant correlation. CASER can provide low-
cost continuous CAD risk monitoring to underserved population of USA including American Indians whose CAD
risk is over 20% the national average.

Grant Number: 1R21HL175632-01A1
NIH Institute/Center: NIH
Principal Investigator: Ayan Banerjee